1 From gene expression to reaction presence

When looking into the UMAPs for scRNAseq data using all genes versus using only the metabolic genes, there is a marked difference in the capacity to separate the cells into their respective cell-types. In fact, only the proliferative T-cells are clearly separated from the rest when using only metabolic genes, which are very well known for their drastic change in metabolism to adjust to their proliferative demands.

Regarding the pseudo-bulk data, which was used to create a cell-type (if present) model for each sample in each individual, we can still see the same separation of the proliferative T-cells. Furthermore, other cell-types seem to be grouping together.

Next, machine learning with a random forest classifier was performed using either pseudo-bulk data or reaction presence data (which represents the reactions that were considered present when constructing the cell-type specific models).

The evaluation metric used to assess the prediction capability was the Mathews correlation coefficient (MCC), as it is appropriate to classify multi-class problems, perfectly symmetric (no class is more important than the other), and not sensitive to class-imbalance (i.e., when the different classes are not evenly represented, which happens in this data, with proliferative T-cells having far less models than other cell-types). The general MCC formula is the following:

\[ MCC = \frac{TP * TN - FP * FN}{\sqrt{(TP + FP)(TP + FN)(TN + FP)(TN + FN)}} \]

The MCC metric varies from -1 to 1. An MCC value 1 means that all cell-types were correctly classified, while a value of -1 means that all cell-types were not well classified. An MCC value of 0 means that the classifier is no better than random guessing.

The dataset that leads to the best results is the pseudo-bulk data, with an MCC greater than 0.75. Prediction capacity slightly decreases when using reaction presence data, with an MCC of around 0.56. This shows that the models do a good job at representing the cell-types at the level of what reactions should be included or not to accurately represent the transcriptomics data.

2 Pathways’ presence

We next checked the percentage of reactions present in each metabolic pathway (relative to all reactions of that pathway in the generic human model) for all models of each cell-type, and visualised the results in an heatmap.

Several metabolic pathways are highly present across all cell-types, namely: aminoacyl-tRNA biosynthesis, arachidonic acid metabolism, estrogen metabolism, fatty acid activation (cytosolic), keratan sulfate biosynthesis, leukotriene metabolism, linoleate metabolism, protein degradation, and vitamin E metabolism.

There are other metabolic pathways that are present somewhat to an equal extent across all cell-types: butanoate metabolism, cholesterol biosynthesis 1 (Bloch pathway), chondroitin / heparan sulfate biosynthesis, glycerolipid metabolism, histidine metabolism, lysine metabolism, porphyrin metabolism, O-glycan metabolism, oxidative phosphorilation, serotonin and melatonin biosynthesis, vitamin D metabolism, and xenobiotics metabolism.

For the remaining pathways, some differences between cell-types can be observed. In fact, overall, memory and regulatory T-cell models have more pthway presence than other cell-types.

Biotin metabolism, for example, has a small reaction presence in IL17+ CD4 T-cells, as well as in proliferative CD4 T-cells, when compared to naive, regulatory, follicular and memory CD4 T-cells. In fact, biotin deficiency was shown to enhance secretion of pro-inflammatory cytokines from CD4 T-cells expressing IL17, as well as differentiation of naive and memory CD4 T-cells toward \(T_{H1}\) and \(T_{H17}\) cells. On the other hand, this deficiency decreases differentiation toward anti-inflammatory regulatory T-cells (Elahi et al. 2018). Finally, biotin deficiency was found to lead to a decrease in CD4 T-cell proliferation (Elahi et al. 2018). In fact, if we plot the presence of biotin metabolism pathway in relation to biomass production in proliferative CD4 T-cell models, we can see a slight positive correlation:

For regulatory CD4 T-cell models, we can see that the reaction presence in some pathways is way lower in models from normal matched tissue than in those from their tumour counterparts. The following heatmap shows this:

In this heatmap, we can see that biotin metabolism is clearly more present in regulatory CD4 T-cell models from tumour tissues. Considering that biotin deficiency decreases differentiation toward anti-inflammatory regulatory T-cells (Elahi et al. 2018), it makes sense that this pathway would be more present in regulatory CD4 T-cells from tumour tissue, where they might be exerting a highly immune suppressive role that helps tumour growth, while in the normal matched, non-inflammatory, tissue it is not as necessary.

The other pathways that have a clear separation between tumour and normal matched mucosa derived models are Biopterin metabolism, Bile acid synthesis, and Fatty acid biosynthesis.

We can see something similar for naive CD8 T-cells, with pathways Glycosylphosphatidylinositol (GPI)-anchor biosynthesis, keratan sulfate degradation, and Beta-oxidation of fatty acids:

However, this time, the metabolic pathways are more present in naive CD8 T-cell models derived from normal matched mucosa tissues than those from tumour tisues.

Fatty acid related pathways are also worth looking into:

Most models from memory CD4 and CD8 T-cells have a high reaction presence of \(\beta\)-oxidation and biosynthesis of FAs, as seen in literature (O’Sullivan et al. (2014)).

Regarding regulatory CD4 T-cells, most models have a high reactions presence of \(\beta\)-oxidation of FAs, also corroborated by the literature (Cluxton, Moran, and Fletcher (2019)). However, regulatory T-cells have been found to not entirely rely in FA oxidation (Cluxton, Moran, and Fletcher (2019), Raud et al. (2018)). This can be the case with those models, all from normal tissue samples, that present low reaction presence of \(\beta\)-oxidation of unsaturated FAs, phytanic acid and even-chain FAs in the peroxisome. These same models also have low reaction presence of odd-chain FA biosynthesis.

Nevertheless, all models from memory CD4 and CD8 and regulatory CD4 T-cells have high reaction presence of FA activation in both cytosol and endoplasmic reticulum.

For models of IL17+ CD4 T-cells, those from normal matched tissue have really low reaction presence of all \(\beta\)-oxidation pathways, while having high reaction presence in the biosynthesis pathway of unsaturated FAs, which is corroborated in the literature (Berod et al. (2014)). On the other hand, models from tumour tissue have high reaction presence of some of the \(\beta\)-oxidation pathways, more specifically oxidation of poly-unsaturated FAs in mitochondria and unsaturated FAs (n-7) in peroxisome.

Regarding naive T-cells, they are characterised by high FA oxidation (MacIver, Michalek, and Rathmell (2013)). In fact, we can see this across most naive CD4 T-cell models, while for naive CD8 T-cell models only those from normal matched tissue show high presence of FA oxidation.

In general, reaction presence of pathways related to fatty acids are in agreement with the literature, with the exception of IL17+ CD4 and naive CD8 T-cell models from tissue samples. This could mean that the tumour micro-environment does affect IL17+ CD4 and naive CD8 T-cells’ FA related metabolism.

3 Flux Prediction under Normal Human Blood Medium

Next, we predicted the reaction fluxes in the models using FBA. For proliferative models, the biomass reaction was optimized, while for the other models both the biomass and the ATP production reactions were optimized. The metabolites included in the medium, and respective fluxes, were based on the Serum Metabolome Database (SMDB), in order to obtain a medium as close as possible to the physiological one.

The biomass and ATP production were checked, as well as the pathways that produce the necessary FADH2 and NADH. The total flux of FAs uptake was also checked.

3.1 Biomass vs ATP production

As expected, proliferative CD4 and CD8 T-cells seem to have a higher biomass flux than their naive counterparts. Regarding the remaining cell-types, each seems to have a varied biomass flux across the different models, with the exception of cytotoxic CD8 and IL17+ CD4 T-cells, which have relatively smaller biomass fluxes.

Naive CD8 T-cells have markedly more ATP production than their proliferative counterparts. However, the same thing for naive vs proliferative CD4 T-cells is not clear.

When comparing biomass and ATP production, ATP production is clearly bigger in IL17+ CD4 T-cells and naive CD8 T-cells. In proliferative CD4 and CD8 T-cells, the flux going through the biomass reaction is clearly bigger than their ATP production.

We then checked how different biomass and ATP production were between cell-type models from normal vs tumour tissue:

Biomass

ATP Production

For cytotoxic CD8 and regulatory CD4 T-cells, models from tumour tissues have higher biomass flux than those from normal matched mucosa. For memory CD8 T-cells, there is increased biomass in models from tumour border.

3.2 Sources of FADH2 and NADH

In general, all cell-types resort obtain most of their FADH2 and NADH from Fatty Acid Oxidation (FAO), which is expected according to the literature.

FADH2

NADH

3.3 FAs Uptake

In general, cytotoxic CD8, naïve CD4 and proliferative CD4 and CD8 T-cells are the cell-types with the least overall uptake of fatty acids (FAs). Although on average the other cell-types uptake more FAs, the uptake is still very varied across the models of each cell-type.

This is in line with the literature. Proliferative T-cells rely more in FA synthesis and decrease FA oxidation relative to their naïve counterparts (Dumitru, Kabat, and Maloy (2018), Kolan et al. (2020)), which might result in decreased necessity for uptake of FAs. Memory CD8 T-cells promote FA oxidation, unlike the activated ones (Kolan et al. (2020)), which might explain why FA uptake for memory CD8 T-cells is generally higher than cytotoxic and proliferative CD8 T-cells.

We then checked how different biomass and ATP production were between cell-type models from normal vs tumour tissue:

4 Medium Changes

We also checked how the models would behave when certain metabolites were excluded from the medium. The exclusion of these metabolites from the medium has been found in previous studies to affect T-cells. As such, the models’ outcomes were compared to those of the literature.

4.1 No Tryptophan

[B2]

It is expected that absence of tryptophan from medium causes T-cells’ biomass to decrease, as it has been shown that indoleamine-pyrrole 2,3-dioxygenase (IDO), which catalyses tryptophan metabolism in the kynurenine pathway, inhibits T-cell activation by tryptophan deprivation and by promoting the expansion of regulatory T-cells (Le Bourgeois et al. (2018)).

Indeed, the biomass of all models, across all cell-types, decreases to zero or very close to zero once tryptophan is removed from medium.

4.2 No Oxygen

[B4, B16]

In general, it is expected that T-cells suffer reduced proliferation in an environment with no oxygen (Dumitru, Kabat, and Maloy (2018), Loeffler, Juneau, and Masserant (1992)), even though reduced amounts of oxygen up-regulate genes involved in glycolytic ATP production and down-regulates the OXPHOS pathways (Tripmacher et al. (2008)), associated with higher (Carswell, Weiss, and Papoutsakis (2000), Haddad et al. (2004)) or no effect (Tripmacher et al. (2008)) in proliferation. Thus, it could be expected that removing oxygen from the metabolic models’ medium would result in decreased biomass. However, that is not the case in any of the cell-types, even though very few models, across most cell-types, do suffer a decrease in biomass flux. A cell-type where no models suffer a decrease in biomass when oxygen is removed from the medium is IL17+ CD4 T-cells. In fact, hypoxia is associated with inducing Th17 cell proliferation (Krzywinska and Stockmann (2018)).

Still, it has been pointed out that the impact of oxygen in cell viability relies on the type of stimulus that the stimulated cultures received, as two different stimuli revealed different impacts of oxygen levels on T-cells proliferation (Atkuri, Herzenberg, and Herzenberg (2005)). This might help explain why some models show a decrease in biomass while others from the same cell-type show the contrary.

One important thing to take in mind is that metabolic models only capture metabolic pathways, i.e., signaling and/or regulation pathways are not captured in this models. This means that the regulation related to the hypoxic inducible factors (HIFs) are not captured in these models.

Finally, as expected, the reaction catalised by oxygen-dependent prolyl hydroxylase domain enzymes (PHDs) does not have any flux in any model when no oxygen is present in the medium:

Next, we checked if the models that uptaked the most amounts of oxygen were the ones that suffered the most in biomass flux when oxygen was no longer available:

What happens to the biomass of the cell-types from normal tissue vs tumour tissue, when there is no oxygen:

4.3 No Glutamine

[B8]

Unavailability if glutamine in medium, with glucose present, decreases biomass flux to or very close to zero across all cell-types.

As reported in the literature, glutamine seems to be essential for all T-cells’ proliferation (Dumitru, Kabat, and Maloy (2018), Kolan et al. (2020), Le Bourgeois et al. (2018)), especially because it acts as a nitrogen donor for DNA and RNA nucleotide production (Dumitru, Kabat, and Maloy (2018), Kolan et al. (2020)). Indeed, DNA and RNA production decreases to zero, or very close to zero, when no glutamine is available in the models’ medium.

[B9]

DNA Production

RNA Production

4.4 No Nucleotides

[B17]

Overall, there is no difference in biomass flux and production of DNA when no nucleotides are available in the medium. This goes in line with a study (Varanasi, Ma, and Kaech (2019)) on in vitro vs in vivo metabolismof CD8+ T-cells, where the effector cells where shown to almost entirely rely on de novo nucleotide biosynthesis.

Biomass

Production of Nucleotides

Production of nucleotides assessed through the DNA production reaction:

5 Gene Essentiality

We tested the capacity of the models to accomplish their objective upon single gene deletions. All genes were deleted individually and the models’ objective predicted using ROOM approach. Genes were considered essential when their deletion causes the model to have a 0 in the objective or be infeasible. Nevertheless, we still assessed if the non-essential genes caused a decrease in the objective’s value in some of our analyses.

The most affected pathways were assessed by looking into the percentage of genes in a pathway that were predicted as essential:

There is a clear group of pathways where most genes are essential for the cells. These pathways are CoA synthesis, heme synthesis, ubiquinone synthesis, protein assembly, phenylalanine metabolism, urea cycle, vitamin B2 metabolism.

Pathways with a considerable amount of essential genes, but not as much as the previous ones, include cholesterol biosynthesis, thiamine metabolism, tryptophan metabolism, sulfur metabolism and glycosphingolipid biosynthesis-globo series.

Next, we plotted how much each pathway is represented within the group of essential genes of a model.

A lot of the models’ essential genes encode for enzymes that catalise transport reactions, followed by reactions from the pathways leukotriene metabolism, sphingolipid metabolism, oxidative phosphorilation, and fatty acid oxidation.

After looking into how the essential genes of the different cell-types overlap (next plot), we can see that the cell-type with the most amount of unique essential genes is the proliferative CD4 T-cells, with 93 genes, followed by follicular CD4 (69), proliferative CD8 (32), regulatory CD4 (24), cytotoxic CD8 (20), naïve CD8 (18), naïve CD4 (3), memory CD8 (1), and memory CD4 (1). IL17+ CD4 T-cells have no unique essential genes.

Next are the tables with the essential genes of each cell-type. Information on the pathway they are part of and if they are essential only for that cell-type is available.

Naive CD4 Tcells

Naive CD8 Tcells

Memory CD4 Tcells

Memory CD8 Tcells

Proliferative CD4 Tcells

Proliferative CD8 Tcells

IL17+ CD4 Tcells

Follicular CD4 Tcells

Regulatory CD4 Tcells

Cytotoxic CD8 Tcells

The genes whose knockout does not turn the model’s objective to zero in more than 90% of the cell-type’s models but still decreases it in more than 90% of the cell-type’s models were organized in the following tables.

Naive CD4 Tcells

Naive CD8 Tcells

Memory CD4 Tcells

Memory CD8 Tcells

Proliferative CD4 Tcells

Proliferative CD8 Tcells

IL17+ CD4 Tcells

Follicular CD4 Tcells

Regulatory CD4 Tcells

Cytotoxic CD8 Tcells

5.1 Comparison with Literature

We were able to find two studies that performed gene essentiality using CRISPR screens on CD8+ T-cells. Zhao et al. (2021) performed gene essentiality in CD8+ T-cells from mice, while Shifrut et al. (2018) performed it in human CD8+ T-cells.

As such, we compared the results from these studies to the essential genes that were predicted for the CD8 T-cell subtypes (naive, memory, cytotoxic and proliferative).

From all the 408 genes predicted as essential in our models, 129 were corroborated by the literature. Of these, 37 were present in both mouse and human studies, while 52 and 40 were present only in human and mouse studies, respectively.

Next, we assessed if the genes reported as essential by the studies but not predicted as such by our models were tested for gene essentiality. These genes include the 160 that were only reported by Zhao et al. (2021), the 306 only reported by Shifrut et al. (2018), and the 59 reported by both studies. Only the genes that, once deleted, catalise reactions that do not have alternative gene(s) that could carry the reaction were tested for gene essentiality.

From the total of genes reported by the studies but not predicted as essential by our models, a total of 104 genes were tested, while the other 421 were not considered for essentiality prediction. From the 59 genes predicted as essential by both studies, only 15 were tested, while the other 44 were not.

Regarding the genes that were not tested for essentiality, several reasons might come into place to explain the discrepancy with the studies: (1) the gene-protein-reaction (GPR) rules are not completely correct; (2) the difference between the T-cells in the studies and those used to create the models; (3) the error that comes into creating the models themselves; (4) or even the fact that a gene deletion might affect the cell behond the metabolism, like signaling and regulatory pathways.

Regarding the genes tested but not predicted as essential, however, their ability to decrease in more than 50% the models’ objective was assessed. Indeed, all the essential genes from studies that were tested but not predicted as essential lead to a decrease in the models’ objective:

6 Glucose Case

When only removing glucose from medium, biomass only decreases slightly in the cell-types follicular CD4, proliferative CD8, and regulatory CD4 T-cells. In fact, most models in these cell-types do not suffer changes in the biomass flux, with only some suffering from biomass flux decrease. This, however, is not corroborated by literature, as it has been reported decreased T-cell proliferation rates in glucose-deficient media (Dumitru, Kabat, and Maloy (2018)).

However, these studies are mainly done in vitro, where not all metabolites present in the blood, for example, are used, and where glucose concentration is significantly higher than phisiological levels. This might affect the observed dependence on glucose by T-cells.

Furthermore, there are enzymes catalised by cells that externally produce glucose from other metabolites in the medium. This is the example of sucrose and glucosylceramide pool (GC pool), which are present in our medium. In fact, we can observe flux going through the reactions that produce glucose from sucrose and GC pool, respectively. In some cell-type models, one can see that this flux increases upon glucose depletion from the medium (follicular CD4 Tcells, naive CD8 Tcells, proliferative CD4 nd CD8, and regulatory Tcells).

Also, it is possible to see with our models that the transport of glucose is mainly done through the reaction that represents GLUT1 transporter.

7 Tumour vs Normal Human Blood Medium

To create the tumour medium, studies with metabolite concentrations from CRC medium were not found. However, 6 studies, through Mass-Spectrometry (MS), calculated the fold changes of the metabolites’ intensities in human blood between normal individuals and tumour patients. We found 84 metabolites reported in these studies that are present in our normal human blood medium. With this, we changed their concentration considering the respective fold changes and thus obtaining the tumour human blood medium.

After running FBA with the new medium, we compared the predictions between the two media for each model. After performing a paired statistical test (Wilcoxon signed rank test), we found out that the flux predictions of 71 of the 172 models (approximately 41% of the models) varied significantly when the medium changed.

Of the 71 models whose predictions changed significantly, 7 were cytotoxic CD8 T-cells (~41% of the cytotoxic CD8 models); 3 follicular CD4 (~25%); 3 IL17+ CD4 (~33%); 13 memory CD4 (~46%); 10 memory CD8 (~40%); 16 naive CD4 (~55%); 2 naive CD8 (~25%); 5 prolifrative CD4 (~46%); 3 proliferative CD8 (~50%); and 9 regulatory CD4 (~33%).

We then took a look into the top 15 pathways most affected by the medium change, as well as the top 15 least affected ones. This was assessed by calculating, on average across all models, the percentage of reactions in a pathway with different fluxes between the two media.

Two pathways did not have any reactions changing in any of the models: dietary fiber binding and CoA metabolism.

Among the most affected pathways are oxidative phosphorilation, pentose phosphate pathway, glycolysis / gluconeogenesis, and pathways related to \(\beta\)-Oxidation of fatty acids and carnithine shuttle.

References

Atkuri, Kondala R, Leonard A Herzenberg, and Leonore A Herzenberg. 2005. “Culturing at Atmospheric Oxygen Levels Impacts Lymphocyte Function.” Proceedings of the National Academy of Sciences 102 (10): 3756–59.
Berod, L., C. Friedrich, A. Nandan, and others. 2014. “De Novo Fatty Acid Synthesis Controls the Fate Between Regulatory T and T Helper 17 Cells.” Nature Medicine 20 (11): 1327.
Carswell, KS, JW Weiss, and ET Papoutsakis. 2000. “Low Oxygen Tension Enhances the Stimulation and Proliferation of Human t Lymphocytes in the Presence of IL-2.” Cytotherapy 2 (1): 25–37.
Cluxton, D., B. Moran, and J. M. Fletcher. 2019. “Differential Regulation of Human Treg and Th17 Cells by Fatty Acid Synthesis and Glycolysis.” Frontiers in Immunology 10: 115.
Dumitru, Cristina, Agnieszka M Kabat, and Kevin J Maloy. 2018. “Metabolic Adaptations of Cd4+ t Cells in Inflammatory Disease.” Frontiers in Immunology 9: 540.
Elahi, Asif, Subrata Sabui, Nell N Narasappa, Sudhanshu Agrawal, Nils W Lambrecht, Anshu Agrawal, and Hamid M Said. 2018. “Biotin Deficiency Induces Th1-and Th17-Mediated Proinflammatory Responses in Human Cd4+ t Lymphocytes via Activation of the mTOR Signaling Pathway.” The Journal of Immunology 200 (8): 2563–70.
Haddad, Hadar, Dirk Windgassen, Christopher G Ramsborg, Carlos J Paredes, and Eleftherios T Papoutsakis. 2004. “Molecular Understanding of Oxygen-Tension and Patient-Variability Effects on Ex Vivo Expanded t Cells.” Biotechnology and Bioengineering 87 (4): 437–50.
Kolan, Shrikant S, Gaoyang Li, Jonas A Wik, Giulia Malachin, Shuai Guo, Pratibha Kolan, and Bjørn S Skålhegg. 2020. “Cellular Metabolism Dictates t Cell Effector Function in Health and Disease.” Scandinavian Journal of Immunology 92 (5): e12956.
Krzywinska, Ewelina, and Christian Stockmann. 2018. “Hypoxia, Metabolism and Immune Cell Function.” Biomedicines 6 (2): 56.
Le Bourgeois, Thibault, Laura Strauss, Halil-Ibrahim Aksoylar, Saeed Daneshmandi, Pankaj Seth, Nikolaos Patsoukis, and Vassiliki A Boussiotis. 2018. “Targeting t Cell Metabolism for Improvement of Cancer Immunotherapy.” Frontiers in Oncology 8: 237.
Loeffler, DA, PL Juneau, and S Masserant. 1992. “Influence of Tumour Physico-Chemical Conditions on Interleukin-2-Stimulated Lymphocyte Proliferation.” British Journal of Cancer 66 (4): 619–22.
MacIver, N. J., R. D. Michalek, and J. C. Rathmell. 2013. “Metabolic Regulation of T Lymphocytes.” Annual Review of Immunology 31: 259–83.
O’Sullivan, D., G. J. W. van der Windt, S. C. C. Huang, and others. 2014. “Memory CD8+ T Cells Use Cell-Intrinsic Lipolysis to Support the Metabolic Programming Necessary for Development.” Immunity 41 (1): 75–88.
Raud, B., D. G. Roy, A. S. Divakaruni, and others. 2018. “Etomoxir Actions on Regulatory and Memory T Cells Are Independent of Cpt1a-Mediated Fatty Acid Oxidation.” Cell Metabolism 28 (3): 504–15.
Shifrut, Eric, Julia Carnevale, Victoria Tobin, Theodore L Roth, Jonathan M Woo, Christina T Bui, P Jonathan Li, Morgan E Diolaiti, Alan Ashworth, and Alexander Marson. 2018. “Genome-Wide CRISPR Screens in Primary Human t Cells Reveal Key Regulators of Immune Function.” Cell 175 (7): 1958–71.
Tripmacher, Robert, Timo Gaber, René Dziurla, Thomas Häupl, Kerem Erekul, Andreas Grützkau, Miriam Tschirschmann, et al. 2008. “Human Cd4+ t Cells Maintain Specific Functions Even Under Conditions of Extremely Restricted ATP Production.” European Journal of Immunology 38 (6): 1631–42.
Varanasi, Siva Karthik, Shixin Ma, and Susan M Kaech. 2019. “T Cell Metabolism in a State of Flux.” Immunity 51 (5): 783–85.
Zhao, Hanfei, Ying Liu, Lixia Wang, Gang Jin, Xiaocui Zhao, Jing Xu, Guangyue Zhang, Yuying Ma, Na Yin, and Min Peng. 2021. “Genome-Wide Fitness Gene Identification Reveals Roquin as a Potent Suppressor of Cd8 t Cell Expansion and Anti-Tumor Immunity.” Cell Reports 37 (10): 110083.